Flow-based neural networks for scalable high resolution imaging of VLBI observations

Abstract

Bayesian sampling methods in very-long-baseline interferometry enable reliable imaging of astronomical sources with a record high angular resolution. However, even after the transition from classical MCMC methods to flow-based neural networks, they remain computationally expensive as the number of recovered features grows. We used WaveletFlow neural network architecture to increase the number of pixels in the recovered image distribution, addressing the scaling issues of previous models. We applied the algorithm to both model data and archive RadioAstron project observations of M87 and ALMA telescope observations of HL Tau. While yielding comparable fidelity of the individual images, the algorithm provides more statistical information compared to CLEAN, which may be crucial for future research on compact astronomical objects such as active galactic nuclei.